Hi.

I am reading the UNIX Time Sharing paper by Ritchie and Thompson. There is a section (Implementation of a shell) that I do not understand.

What I get from it is that the shell is a process that keeps reading for new commands. Once it has received one it spawns a new process (using fork()) and that new process calls execute. So far so good. However, it also states that if an argument with > or < is given, then the offspring changes the standard file descriptors (0,1) to refer to the named file, by closing the necessary file (0 or 1) and opening the new file instead. Since both processes will share the file descriptor table, this should work.

However, this is what I don't understand: Once the son has closed a file and opened the new file (for example a command executed by the shell), how will the shell reopen the file descriptor for the input,output files that it previously had? It says here that "the shell need not know the actual names of the files (...) because it need never reopen them". However, why should it not need to reopen them if it closed them before executing the child and needs to keep accessing those files to continue running commands?

Thank you very much!

After fork() there are two processes, each with separate descriptor table. So if one copy closes some of the descriptors, the other one is unaffected. It doesn't need to reopen them, because they have never been closed from its perspective.

Note also the difference between file descriptors ie I/O channel nums (0=stdin, 1=sdtout, 2=stderr) and actual filenames.
0,1,2 are default setup for each shell as it starts, but you can use higher nums to open other streams/files at the same time, iirc at least as high as 9.
This is a good page: http://www.tldp.org/LDP/abs/html/io-redirection.html
My apologies if you already knew this stuff, but might be handy for others.